portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Lane by Lane operation

Browse Source
This commit is contained in:
Peter Boyle 2018-05-12 17:59:35 -04:00
parent d3f51065c2
commit 09cd46d337
1 changed files with 37 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
lib/tensors/Tensor_extract_merge.h
View File

@ -107,6 +107,43 @@ void merge(vobj &vec,ExtractBuffer<sobj> &extracted)
} }
} }
////////////////////////////////////////////////////////////////////////
// Extract/Insert a single lane
////////////////////////////////////////////////////////////////////////
template<class vobj> accelerator_inline
typename vobj::scalar_object extractLane(int lane, const vobj & __restrict__ vec)
{
typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
constexpr int words=sizeof(vobj)/sizeof(vector_type);
constexpr int Nsimd=vector_type::Nsimd();
scalar_object extracted;
scalar_type * __restrict__ sp = (scalar_type *)&extracted; // Type pun
scalar_type * __restrict__ vp = (scalar_type *)&vec;
for(int w=0;w<words;w++){
sp[w]=vp[w*Nsimd+lane];
}
return extracted;
}
template<class vobj> accelerator_inline
void insertLane(int lane, vobj & __restrict__ vec,const typename vobj::scalar_object & __restrict__ extracted)
{
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
constexpr int words=sizeof(vobj)/sizeof(vector_type);
constexpr int Nsimd=vector_type::Nsimd();
scalar_type * __restrict__ sp = (scalar_type *)&extracted;
scalar_type * __restrict__ vp = (scalar_type *)&vec;
for(int w=0;w<words;w++){
vp[w*Nsimd+lane]=sp[w];
}
}
//////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////
// Extract to a bunch of scalar object pointers of different scalar type, with offset. Useful for precision change // Extract to a bunch of scalar object pointers of different scalar type, with offset. Useful for precision change
//////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////